White House Accuses Chinese AI Firm Moonshot of Using Banned Nvidia Chips

A top White House official has accused Chinese artificial intelligence startup Moonshot of illicitly accessing cutting-edge U.S. technology, alleging the firm used banned Nvidia hardware to build its high-performance Kimi K3 model. The accusation raises fresh concerns about the effectiveness of U.S. restrictions designed to slow China's AI development.

Details of the Accusation

Michael Kratsios, Director of the White House Office of Science and Technology Policy, stated that Moonshot acquired servers equipped with Nvidia GB300 chips and routed computing workloads through infrastructure in Thailand to train its models. According to Kratsios, these actions violate both U.S. export regulations and the terms of service of American AI developers whose models were also allegedly used in the process.

The GB300 platform is part of Nvidia's latest Blackwell generation of semiconductors. The U.S. government has explicitly barred the export of this high-end hardware to Chinese companies, though it has granted exemptions for lower-performance processors like the H200.

Regulatory Scrutiny

Federal investigators at the Bureau of Industry and Security (BIS), the agency responsible for enforcing technology transfer rules, are reportedly examining whether Chinese firms are bypassing restrictions by using overseas subsidiaries or cloud infrastructure in third-party countries. This case highlights the challenge regulators face in policing the use of advanced technology across global supply chains.

Ad

Nvidia has previously stated that it complies with all export control regulations and works to ensure compliance among its customers. The company did not immediately comment on the White House's latest statement, according to the source report.

Context: Kimi K3's Advanced Capabilities

The dispute follows the recent release of Moonshot's Kimi K3, an open-weight AI model that has impressed industry analysts with its capabilities. Moonshot claims the model trails only top-tier U.S. systems from OpenAI and Anthropic in benchmark performance, narrowing a previously wide gap between Chinese and American AI labs.

Kratsios asserted that Moonshot achieved these results in part through model distillation, a technique where a smaller "student" model is trained using the outputs of a more advanced "teacher" model. In this case, Moonshot is accused of using proprietary U.S. systems as the teacher models to rapidly advance its own technology.